Compressive Sensing Based PPG Data Reduction

Algorithm

Page | 1
 CONTENTS

Abstract 3
1 Introduction 4
2 Aim 4
3 Project Objective 4
4 Compressive Sensing 5
4.a Design of Dictionaries 7
4.b Sensing Matrix 8
4.c Reconstruction Algorithms 8
4.d Performance Evaluation Metrics 9
5 Related Works 10
6 Result Analysis 11
7 Conclusion 27
8 Reference 28
9 Acknowledgement 30

Page | 2
 Abstract
Compressive sensing (CS) has emerged as an area that opens new perspectives in signal
acquisition and processing. It appears as an alternative to the traditional sampling theory,
endeavouring to reduce the required number of samples for successful signal
reconstruction. In practice, compressive sensing aims to provide saving in sensing resources,
transmission, and storage capacities and to facilitate signal processing in the circumstances
when certain data are unavailable. To that end, compressive sensing relies on the
mathematical algorithms solving the problem of data reconstruction from a greatly reduced
number of measurements by exploring the properties of sparsity and incoherence.
Therefore, this concept includes the optimization procedures aiming to provide the sparsest
solution in a suitable representation domain.

This study investigates and implements the CS based photoplethysmogram (PPG) data
reduction algorithms. The performance of CS based data reduction algorithm for different
combinations of representation techniques is studied. The Discrete cosine transform (DCT)
based PPG compression algorithm is also implemented and compared with the CS based
PPG data reduction algorithms. The performance of CS algorithm is evaluated by
implementing in the Arduino computing platform. In the CS method, the PPG signal is
reconstructed with reduced number of samples by using the Dictionary matrix, including
discrete impulse dictionary (I), cosine dictionary (C), sine dictionary (S) and the ℓ1-norm
optimisation algorithm. To justify the selection of Dictionary matrix, compression methods
are implemented using the decomposition techniques based on the dictionaries Ψ = *C+, Ψ =
*I C+ and Ψ = *ICS+. The performance of each method is tested and validated using PPG
databases. The results show that the method with dictionary Ψ = *ICS+ is capable of
effectively compressing the PPG. The CR and PRD metrics are used for assessing the quality
of the reconstructed signal. From the results it is observed that the CS method had higher
CR with PRD value of below 10 clearly indicate that the proposed Dictionary matrix Ψ= *I C
S] is capable of reproducing the original.

Page | 3
 1 Introduction

The fundamental approach for signal reconstruction from its measurements is defined by
the Shannon-Nyquist sampling theorem stating that the sampling rate needs to be at least
twice the maximal signal frequency [1]. In the discrete case, the number of measurements
should be at least equal to the signal length in order to be exactly reconstructed. However,
this approach may require large storage space, significant sensing time, heavy power
consumption, and large number of sensors. Compressive sensing (CS) is a novel theory that
goes beyond the traditional approach. It shows that a sparse signal can be reconstructed
from much fewer incoherent measurements. The basic assumption in CS approach is that
most of the signals in real applications have a concise representation in a certain transform
domain where only few of them are significant, while the rest are zero or negligible. This
requirement is defined as signal sparsity. Another important requirement is the incoherent
nature of measurements (observations) in the signal acquisition domain. Therefore, the
main objective of CS is to provide an estimate of the original signal from a small number of
linear incoherent measurements by exploiting the sparsity property [2]. The CS theory
covers not only the signal acquisition strategy, but also the signal reconstruction possibilities
and different algorithms

2 Aim

Aim of this project work is to investigate the performance of CS based data reduction
algorithm and DCT based data reduction algorithms or methods for transmission of PPG
signal.

3 Project Objective

 Survey of CS based PPG data reduction algorithm.

 Implementation of CS based PPG data reduction algorithm.
 Investigating performance of CS based data reduction algorithm for different
combination of representation techniques.
 Implementing DCT based PPG compression algorithm.
 Implementing CS algorithm in Arduino computing platform

Page | 4
 4 Compressive Sensing

CS is a two-stage paradigm that acquires the data in a compressed form at the sensor and
reconstructs efficiently the original data from a fewer sample signal at the receiver. Thus, CS
simultaneous senses and compresses the data in one single operation. The well-established
traditional sensing and processing theory relies mainly on the Nyquist–Shannon sampling
theorem which presents the main pillar of the digital processing realm. This theorem states
that, given any signal with a bandwidth of W, its information is entirely preserved if it is
sampled at secured sampling frequency fs ≥ 2W. Further, with the ever-increasing number
of digital sensing devices, several challenges have faced this paradigm, from the torrent
amount of data generated by the different sensing devices to the high data transmission
rate and unbearable constraints on the sampling/processing platforms. Such challenges
would largely increase the cost of the wireless transmission and shorten the sensors
lifespan. Thus, to address the challenges associated with high-dimensional data, researches
depend often on compression techniques by discarding the redundant samples in the signal.
The benefits of compression will apply to the whole ubiquitous computing environment, as
decreasing data size means shorter communication delays, efficient usage of bandwidth and
reduced battery drain[5].
In digital CS theory, any sparse or compressible signal can be given as
∑ (1)

Where is the n x 1 column vector of weighting coefficients and T denotes the

transpose operation. is the basis matrix of size N x N. Therefore, the compressed signal y is
found as
(2)
The Pictorial depiction of (1) and (2) is shown in Fig 4(a), 4(b) and 4(c)

𝓍 𝜓 𝛼

Fig 4 (a): Sparse representation of signal

Page | 5
 𝑦 𝛷 𝔁

Fig 4 (b): CS measurement process

𝑦 𝛷 𝜓 𝛼

Fig 4 (c): Measurement process interms of product

and D have two interesting properties, i.e. they are incoherent with the basis [ ] and they
have the Restricted Isometry Property (RIP) with accurate level for detection probability of
the compressed signals at the receiver side that is suitable for recovering the actual signal
from compressed signal. CS scenario has two important steps
 A stable measurement matrix [ ]M×N should confirm that the significant data in any
compressible data is not dented by the dimensionality drop from down to

 The CS theory offers a restoration algorithm under certain conditions with adequate
exactness to get back actual data x from the compressed data
Therefore, CS theory mainly focus on M random linear measurements instead of N samples
such that M<< N. The compressive sensing theory also offers a reconstruction algorithm to
recover original signal from the compressed signal y only with M random linear
measurements. The number of random linear measurements M, number of
samples/coefficients N, and non-zero significant coefficients K must satisfy the following
equation
M ≥ c K log(N/K) (3)

Page | 6
where c is constant. The CS theory consists of the following steps :
1. Design a stable measurements matrix .
2. Develop a rebuilding algorithm to get back the actual signal from the compressed signal
in the receiver side.
3. Test the number of M measurement to ensure that reduction from N to M is not
damaged to collect the original signal
CS theory offers a way to reconstruct the compressible or sparse signal from the M linear
measurements. There reconstruction algorithm needs only random matrix and to
recover the original signal. norm is used to recover the sparse solution of the signal in
order to reconstruct the signal, where p is a norm parameter. With p = 1 is the of norm,
by minimizing this norm one can reconstruct the sparse solution of the signal with enough
accuracy by solving the basis pursuit denoise problem.

4.a Design of Dictionaries

Choosing an optimal overcomplete matrix is appealing because it leads to simple and fast
algorithms for the sparse representations of signals. In this work, the pre-defined Dictionary
matrix includes cosine dictionary, impulse dictionary and sinusoidal dictionary. Compared
with a learned dictionary approach, such a pre-defined dictionary matrix has the potential
to provide adequate sparse representations of signals. Here, the dictionarymatrix of size
N ×M is constructed as

[ ] [ | | ]

where I denotes impulse matrix of size N × N, C denotes the discrete cosine matrix of size
N × L and S denotes the discrete sine matrix of size N × L. The impulse dictionary consists of
a set of elementary impulse waveforms that are time-shifted discrete impulse functions. The
impulse dictionary matrix is constructed as

The cosine dictionary (or DCT matrix) consists of a set of elementary waveforms that are
sampled cosine functions. The elementary waveforms are column vectors of the cosine
dictionary, which is constructed as

Page | 7
where √ for i=0, otherwise = 1. The sine dictionary (or DST matrix) consists of a
set of elementary waveforms that are sampled sine functions. The elementary waveforms
are column vectors of the sine dictionary, which is constructed as

where for i=L-1, otherwise = 1. We use both discrete cosine and sine
√
dictionaries in order to reduce the waveform discontinuities at the block boundaries, and to
estimate the power quantities directly from the approximation coefficients for the
deterministic sinusoidal components. Depending on PQ disturbance detection, classification,
parameter estimation, compression and denoising problems, the sizes of the cosine and sine
dictionaries can be adopted based on the frequency characteristics of the signals of
interests[1].

4.b Sensing Matrix

Ideally, the sensing matrix has to fulfill the following constraints

 Optimal or near-optimal reconstruction performance

 Optimized sensing performance
 Universality
 Low complexity, fast computation and structure-based processing
 Hardware friendly

4.c Reconstruction Algorithms

In compressed sensing, the reconstruction quality of a signal is not only related to the
selection of the measurement matrix but also to the reconstruction algorithm of the signal.
A good reconstruction algorithm can more closely approximate the original signal and
reduce the error before and after reconstruction [8]. Fast and efficient reconstruction
algorithms are the keys to incorporate CS in real-world applications, thus developing such
algorithms have been the main concern in the CS community.

The sparsity of the signal can be defined as a number of nonzero elements within a vector. It
can be described by using the norms, but it is not feasible in practice because small noise
in the signal will be assumed as a non-zero sample. Therefore, the norms is commonly
used. The optimization problem based on the ɭ1 norms is recast as follows

Page | 8
 ‖ ‖

4.d Performance Evaluation Metrics

In the literature, two performance metrics are used to evaluate the performance of the
lossy compression methods. The compression ratio (CR) is defined as

where No denotes the number of bits used for the original signal and Nc denotes the number
of bits used for the compressed signal. The higher CR indicates the better data compression.
In the literature, different objective signal quality metrics, including percent root-mean-
square difference (PRD), root-mean-square error (RMSE), mean square error (MSE) and
signal-to-noise ratio (SNR) are used for measuring the signal distortion between the original
signal and the reconstructed signal. The SNR is widely used for the performance comparison
of various methods. The SNR is defined as

∑ ‖ ‖
∑ ‖ ̂ ‖

where x denotes the original signal, ̂ denotes the reconstructed signal and N denotes the
length of the signal block. The larger SNR value indicates the higher quality of the
reconstructed signal. It is well known that the signal distortion metrics (or quality
assessment metrics) such as MSE, RMSE, PRD, and SNR distribute the reconstruction error
equally over all samples of the PQ signal.

The use of the percentage root-mean-square difference (PRD) constraint helps recover the
entire signal.

∑ ̂
√
∑

Where is the original signal and ̂ is the reconstructed signal. N is the length (i.e.
dimensionality) of both the signals.[6]

Root mean squared error (RMSE) were also used to evaluate the quality of the
reconstructed signal.

√ ∑ ̂

where x is the original signal and xˆ is the reconstructed signal[8].

Page | 9
 5 Related Works

Candes, Tao, Romberg, and Donoho have turned the world upside down. They proposed the
compressed sensing concept in [4] to capture and represent compressible signals at a rate
significantly below the Nyquist rate. The non-adaptability in linear projections that preserve
the structure of the signal and the signal reconstruction from these projections using an
optimization process are elaborated in these articles. The number of measurements in the
sampling is assumed to be smaller than the length of the signal [5]. Moreover in the
reconstruction, a good approximation to the signal or its coefficients is extracted from the
measurements by solving linear programs such as Basis Pursuit or Matching pursuit
algorithms[6].
If the objects of interest are sparse in a fixed basis or compressible, then it is possible to
reconstruct the signal with very high accuracy from a small number of random
measurements by solving a simple linear program. Candes et al... in [11] proposed methods
for recovering a finite signal from a limited set of measurements
In [11], Candes along with his group solved the problem for recovering an unknown sparse
signal. Proposed a solution called norm for imperfect scenarios where the
measurements are noisy and the signal is not exactly sparse. The methodology also explains
why one can also very nearly recover approximately sparse signals.
Demonstrated theoretically and empirically that a greedy algorithm called Orthogonal
Matching Pursuit (OMP) can reliably recover a signal with m nonzero entries in dimension d
from random linear measurements of that signal[5]. These works proved that when the
signals are not sparse, the non-optimal columns of the matrix are statistically correlated
with the residual vectors generated by the imaginary invocation of the algorithm. It is stated
that the OMP is an effective alternative to Basis Pursuits for signal recovery from random
measurements and also analyzed its performance with greedy pursuit algorithm.
A new iterative recovery algorithm called CoSaMP which delivers the same guarantees as
the best optimization-based approaches. Moreover, this algorithm offers rigorous bounds
on computational cost and storage [15]. It is likely to be extremely efficient for practical
problems because it requires only matrix vector multiplies with the sampling matrix.

Page | 10
 6 Result Analysis

The CS based compression algorithm has been implemented and tested using PPG signal
from MIMIC database, slp database and ourdatabase. CS algorithm were tested by
simulation using MATLAB 2014a. The 10-bit PPG data were collected as a sample set for
dictionary construction, and the whole sample sets were segmented into window/block size
of length N=512 samples from the MIMIC database with sampling rates Fs = 125 Hz, slp
database with sampling frequency Fs = 250 Hz, ourdatabase with sampling frequency Fs =
250 Hz to estimate ̅ . Select the signals from the database and resampled it to 250
Hz. After training, the trained redundant dictionary D was obtained as an experimental
dictionary. For quantization the bits per measurement set as BM=10 bits. In general high
resolution PPG data (10 bit) is preferable over conventional 8-bit for better diagnosis and
meeting accuracy requirements but needs evaluation. For 512 samples the compressive
measurements taken as M=64,128 and 256. The proposed method is implemented using
MATLAB software and the results of this study are summarized in Table 1 to 9 for three
dictionary basis cosine, sine cosine, and sine cosine identity basis.

For dictionary cosine basis, sine cosine, and sine cosine identity basis the method has higher
CR for the number of measurement M = 64. The Table 1, Table 4 and Table7 show the
performance of CS method with dictionary constructed by using cosine basis and different
values of measurements. The Table 2, Table 5 and Table 8 show the performance of CS
method with dictionary constructed by using sine cosine basis and different values of
measurements. The Table 3, Table 6 and Table 9 show the performance of CS method with
dictionary constructed by using sine cosine identity basis and different values of
measurements.

The CR and PRD metrics are used for assessing the quality of the reconstructed signal. From
the results it is observed that the CS method had higher CR for M=64 with PRD value of
below 10 clearly indicate that the proposed OHD matrix Ψ= *I C S+ is capable of reproducing
the original.

Table 1 Performance of MIMIC database for measurements of M =64, 128, and 256 for cosine dictionary.

M CR λ Rel. Tot RMSE SNR PRD MSE

9.6 0.01 0.01 0.2442 6.4258 47.7211 0.8691
9.6 0.01 0.01 0.107 12.2519 24.4008 0.2981
9.6 0.01 0.01 0.1044 11.6789 26.0649 0.2766
9.6 0.01 0.01 0.1463 8.2602 38.636 0.6053
9.6 0.01 0.01 0.1775 8.0639 39.5191 0.5928
64
9.6 0.01 0.01 0.2336 6.399 47.8683 0.7944
9.6 0.01 0.01 0.133 10.4492 30.0288 0.3869
9.6 0.01 0.01 0.11 11.6361 26.1936 0.3153
9.6 0.01 0.01 0.1466 5.1749 55.113 1.0246
9.6 0.01 0.01 0.1616 10.2049 30.8857 0.8578

Page | 11
 9.6 0.01 0.01 0.1301 11.0224 28.1114 0.4432
9.6 0.01 0.01 0.0878 14.3764 19.1065 0.3959
9.6 0.01 0.01 0.1401 10.177 30.9848 0.5114
9.6 0.01 0.01 0.2254 6.8924 45.2253 0.7296
9.6 0.01 0.01 0.2431 5.8438 51.0281 0.7421
9.6 0.01 0.01 0.1111 12.7904 22.934 0.3705
9.6 0.01 0.01 0.1242 10.9561 28.3267 0.4424
9.6 0.01 0.01 0.1823 8.3203 38.3693 0.5715
9.6 0.01 0.01 0.1849 8.6152 37.0886 0.6949
9.6 0.01 0.01 0.1377 10.8928 28.5338 0.8486
4.8 0.01 0.01 0.1448 10.966 28.2945 0.899
4.8 0.01 0.01 0.0475 19.3063 10.8315 0.3017
4.8 0.01 0.01 0.0437 19.2421 10.9117 0.2095
4.8 0.01 0.01 0.0252 23.5201 6.668 0.2286
4.8 0.01 0.01 0.0764 15.3792 17.0231 0.536
4.8 0.01 0.01 0.0589 18.3679 12.0672 0.3654
4.8 0.01 0.01 0.0569 17.8317 12.8356 0.3519
4.8 0.01 0.01 0.0631 16.4551 15.04 0.2672
4.8 0.01 0.01 0.0915 9.2645 34.4172 0.9596
4.8 0.01 0.01 0.1142 13.2191 21.8296 0.5649
128
4.8 0.01 0.01 0.0383 21.6341 8.285 0.2651
4.8 0.01 0.01 0.0402 21.1576 8.7522 0.3272
4.8 0.01 0.01 0.0603 17.5005 13.3345 0.3989
4.8 0.01 0.01 0.0985 14.0823 19.7645 0.6444
4.8 0.01 0.01 0.049 19.7571 10.2836 0.2176
4.8 0.01 0.01 0.028 24.7549 5.7844 0.189
4.8 0.01 0.01 0.0665 16.3854 15.161 0.2875
4.8 0.01 0.01 0.0454 20.3991 9.551 0.2866
4.8 0.01 0.01 0.1037 13.636 20.8065 0.5291
4.8 0.01 0.01 0.078 15.8276 16.1667 0.674
2.4 0.01 0.01 0.068 17.5261 13.2952 0.5237
2.4 0.01 0.01 0.0239 25.2611 5.4569 0.1938
2.4 0.01 0.01 0.0265 23.5705 6.6294 0.1516
2.4 0.01 0.01 0.0171 26.8912 4.5231 0.2008
2.4 0.01 0.01 0.0458 19.8226 10.2064 0.3015
2.4 0.01 0.01 0.0221 26.8786 4.5297 0.1
2.4 0.01 0.01 0.0262 24.5506 5.922 0.1648
2.4 0.01 0.01 0.0268 23.9035 6.3801 0.186
2.4 0.01 0.01 0.0738 11.1346 27.7504 0.6918
2.4 0.01 0.01 0.0517 20.109 9.8753 0.3916
256
2.4 0.01 0.01 0.0219 26.4796 4.7427 0.1813
2.4 0.01 0.01 0.025 25.2977 5.4339 0.2532
2.4 0.01 0.01 0.0337 22.5566 7.4502 0.2273
2.4 0.01 0.01 0.0483 20.27 9.694 0.3819
2.4 0.01 0.01 0.0207 27.2442 4.343 0.1246
2.4 0.01 0.01 0.0192 28.0489 3.9587 0.1811
2.4 0.01 0.01 0.0296 23.402 6.7592 0.1752
2.4 0.01 0.01 0.026 25.2254 5.4793 0.181
2.4 0.01 0.01 0.0477 20.3853 9.5661 0.3711
2.4 0.01 0.01 0.0366 22.3916 7.5931 0.3274

Table 2 Performance of MIMIC database for measurements of M =64, 128, and 256 for sine and cosine
dictionary.
M CR λ Rel. Tot RMSE SNR PRD MSE
9.6 0.01 0.01 0.0932 14.7957 18.206 0.3498
9.6 0.01 0.01 0.0959 13.2033 21.8692 0.3313
9.6 0.01 0.01 0.0591 16.614 14.7673 0.3264
9.6 0.01 0.01 0.0906 12.4181 23.9384 0.5548
9.6 0.01 0.01 0.1088 12.31 24.2382 0.4431
9.6 0.01 0.01 0.1111 12.8594 22.7525 0.4349
64 9.6 0.01 0.01 0.0469 19.5119 10.5781 0.2839
9.6 0.01 0.01 0.0635 16.4067 15.1239 0.2682
9.6 0.01 0.01 0.1144 7.3239 43.0332 1.2179
9.6 0.01 0.01 0.0547 19.6164 10.4515 0.3257
9.6 0.01 0.01 0.0669 16.7964 14.4604 0.4346
9.6 0.01 0.01 0.0421 20.7659 9.156 0.3982
9.6 0.01 0.01 0.0423 20.5769 9.3574 0.342

Page | 12
 9.6 0.01 0.01 0.0731 16.6722 14.6686 0.3703
9.6 0.01 0.01 0.0814 15.3469 17.0866 0.3007
9.6 0.01 0.01 0.0554 18.8364 11.4335 0.3528
9.6 0.01 0.01 0.0443 19.912 10.1019 0.422
9.6 0.01 0.01 0.0579 18.2758 12.1958 0.4272
9.6 0.01 0.01 0.069 17.1714 13.8494 0.5287
9.6 0.01 0.01 0.067 17.1486 13.8857 0.738
4.8 0.01 0.01 0.044 21.3119 8.5982 0.3787
4.8 0.01 0.01 0.0327 22.5562 7.4506 0.3953
4.8 0.01 0.01 0.0188 26.5509 4.7039 0.2742
4.8 0.01 0.01 0.026 23.2786 6.856 0.3167
4.8 0.01 0.01 0.0375 21.5682 8.3481 0.3708
4.8 0.01 0.01 0.0303 24.1531 6.1994 0.2692
4.8 0.01 0.01 0.0294 23.5606 6.637 0.2977
4.8 0.01 0.01 0.0302 22.8666 7.189 0.2243
4.8 0.01 0.01 0.0817 10.2515 30.7203 1.2278
4.8 0.01 0.01 0.0488 20.607 9.325 0.3295
128
4.8 0.01 0.01 0.0335 22.8009 7.2436 0.3396
4.8 0.01 0.01 0.0293 23.9064 6.3779 0.3501
4.8 0.01 0.01 0.0246 25.2775 5.4466 0.3083
4.8 0.01 0.01 0.0341 23.2998 6.8393 0.3025
4.8 0.01 0.01 0.0231 26.2726 4.857 0.2758
4.8 0.01 0.01 0.0244 25.9377 5.0479 0.2774
4.8 0.01 0.01 0.0309 23.038 7.0485 0.3926
4.8 0.01 0.01 0.0328 23.2145 6.9068 0.4322
4.8 0.01 0.01 0.0405 21.8034 8.1252 0.4822
4.8 0.01 0.01 0.0492 19.8349 10.1919 0.6906
2.4 0.01 0.01 0.0291 24.8901 5.695 0.3928
2.4 0.01 0.01 0.024 25.2218 5.4816 0.3761
2.4 0.01 0.01 0.0161 27.9025 4.026 0.2648
2.4 0.01 0.01 0.016 27.5029 4.2156 0.2213
2.4 0.01 0.01 0.0239 25.4723 5.3258 0.373
2.4 0.01 0.01 0.0154 30.0208 3.1547 0.1795
2.4 0.01 0.01 0.0182 27.7082 4.1171 0.2644
2.4 0.01 0.01 0.0163 28.197 3.8918 0.1757
2.4 0.01 0.01 0.074 11.1057 27.8428 1.2448
2.4 0.01 0.01 0.0336 23.8388 6.4277 0.2583
256
2.4 0.01 0.01 0.0169 28.7624 3.6465 0.2144
2.4 0.01 0.01 0.0188 27.7707 4.0876 0.2359
2.4 0.01 0.01 0.0175 28.2454 3.8702 0.2724
2.4 0.01 0.01 0.019 28.3961 3.8036 0.1958
2.4 0.01 0.01 0.0174 28.7701 3.6433 0.2647
2.4 0.01 0.01 0.0182 28.4823 3.7661 0.1974
2.4 0.01 0.01 0.0229 25.6361 5.2263 0.3749
2.4 0.01 0.01 0.0257 25.3411 5.4069 0.4247
2.4 0.01 0.01 0.0293 24.6213 5.874 0.4406
2.4 0.01 0.01 0.033 23.2883 6.8483 0.5561

Table 3 Performance of MIMIC database for measurements of M =64, 128, and 256 for sine, cosine and
identity dictionary.
M CR λ Rel. Tot RMSE SNR PRD MSE
9.6 0.01 0.01 0.1142 13.0313 22.3066 0.3273
9.6 0.01 0.01 0.1142 13.0313 22.3066 0.3273
9.6 0.01 0.01 0.112 11.8549 25.5421 0.3418
9.6 0.01 0.01 0.0821 13.7668 20.4956 0.3649
9.6 0.01 0.01 0.1041 11.2156 27.4928 0.5449
9.6 0.01 0.01 0.135 10.4376 30.0692 0.425
9.6 0.01 0.01 0.1609 9.6377 32.9699 0.5356
9.6 0.01 0.01 0.0632 16.9124 14.2686 0.2617
64
9.6 0.01 0.01 0.0735 15.1326 17.5133 0.2895
9.6 0.01 0.01 0.1 8.4986 37.5899 1.2007
9.6 0.01 0.01 0.0742 16.9708 14.173 0.3842
9.6 0.01 0.01 0.0752 15.7841 16.2478 0.449
9.6 0.01 0.01 0.054 18.6083 11.7378 0.441
9.6 0.01 0.01 0.0663 16.6714 14.6701 0.3717
9.6 0.01 0.01 0.1011 13.8548 20.2889 0.4219
9.6 0.01 0.01 0.0975 13.7774 20.4706 0.2776

Page | 13
 9.6 0.01 0.01 0.0688 16.9499 14.2071 0.3641
9.6 0.01 0.01 0.0578 17.5984 13.185 0.4483
9.6 0.01 0.01 0.0684 16.8391 14.3894 0.4398
9.6 0.01 0.01 0.0724 16.7531 14.5326 0.5445
4.8 0.01 0.01 0.0546 19.4319 10.6759 0.3763
4.8 0.01 0.01 0.0546 19.4319 10.6759 0.3763
4.8 0.01 0.01 0.0331 22.4507 7.5416 0.4172
4.8 0.01 0.01 0.0258 23.8148 6.4455 0.2929
4.8 0.01 0.01 0.0381 19.9567 10.05 0.4373
4.8 0.01 0.01 0.0431 21.0852 8.8255 0.3582
4.8 0.01 0.01 0.034 22.2989 7.6745 0.3108
4.8 0.01 0.01 0.0473 18.9663 11.2638 0.288
4.8 0.01 0.01 0.0714 11.4243 26.8403 1.1646
4.8 0.01 0.01 0.046 21.1301 8.78 0.3596
128
4.8 0.01 0.01 0.0405 21.1493 8.7606 0.4063
4.8 0.01 0.01 0.0413 20.9214 8.9935 0.4257
4.8 0.01 0.01 0.0342 22.4166 7.5713 0.3322
4.8 0.01 0.01 0.0453 20.8224 9.0966 0.3487
4.8 0.01 0.01 0.0345 22.8004 7.244 0.2981
4.8 0.01 0.01 0.0373 22.2703 7.6999 0.3802
4.8 0.01 0.01 0.0387 21.0847 8.826 0.4231
4.8 0.01 0.01 0.0406 21.3532 8.5574 0.4645
4.8 0.01 0.01 0.0495 20.0603 9.9309 0.5398
4.8 0.01 0.01 0.0558 18.7415 11.5591 0.7839
2.4 0.01 0.01 0.039 22.3716 7.6107 0.3909
2.4 0.01 0.01 0.039 22.3716 7.6107 0.3909
2.4 0.01 0.01 0.0437 20.0264 9.9696 0.4129
2.4 0.01 0.01 0.0308 22.2812 7.6903 0.28
2.4 0.01 0.01 0.0337 21.0241 8.8878 0.4105
2.4 0.01 0.01 0.0354 22.0621 7.8867 0.4331
2.4 0.01 0.01 0.0373 22.3374 7.6406 0.3289
2.4 0.01 0.01 0.0287 23.7695 6.4793 0.2951
2.4 0.01 0.01 0.0339 21.8611 8.0713 0.2791
2.4 0.01 0.01 0.0686 11.7683 25.7979 1.1803
256
2.4 0.01 0.01 0.0423 21.8463 8.0851 0.3422
2.4 0.01 0.01 0.0378 21.7589 8.1669 0.3924
2.4 0.01 0.01 0.0386 21.5111 8.4032 0.4157
2.4 0.01 0.01 0.032 23.0123 7.0694 0.321
2.4 0.01 0.01 0.039 22.3716 7.6107 0.3909
2.4 0.01 0.01 0.039 22.3716 7.6107 0.3909
2.4 0.01 0.01 0.0437 20.0264 9.9696 0.4129
2.4 0.01 0.01 0.0308 22.2812 7.6903 0.28
2.4 0.01 0.01 0.0337 21.0241 8.8878 0.4105
2.4 0.01 0.01 0.0354 22.0621 7.8867 0.4331

Table 4 Performance of slp database for measurements of M =64, 128, and 256 for cosine dictionary.
M CR λ Rel. Tot RMSE SNR PRD MSE
9.6 0.01 0.01 0.0997 11.1095 27.8307 0.2921
9.6 0.01 0.01 0.2198 6.0182 50.014 1.0077
9.6 0.01 0.01 0.2532 3.7862 64.6679 1.0616
9.6 0.01 0.01 0.0963 15.1828 17.4125 0.2896
9.6 0.01 0.01 0.0776 13.5071 21.1175 0.2791
9.6 0.01 0.01 0.1314 11.7392 25.8844 0.3939
9.6 0.01 0.01 0.1067 11.5878 26.3396 0.5171
9.6 0.01 0.01 0.2139 5.8117 51.2173 0.877
64 9.6 0.01 0.01 0.0921 16.8379 14.3915 0.5412
9.6 0.01 0.01 0.2099 6.2186 48.8729 0.6747
9.6 0.01 0.01 0.1085 12.5565 23.56 0.3731
9.6 0.01 0.01 0.1244 8.8574 36.0685 0.4324
9.6 0.01 0.01 0.105 12.8677 22.7309 0.3699
9.6 0.01 0.01 0.0799 14.3949 19.0659 0.2673
9.6 0.01 0.01 0.0764 15.4948 16.7982 0.3637
9.6 0.01 0.01 0.2048 6.9446 44.9542 0.774
9.6 0.01 0.01 0.0545 19.3284 10.8038 0.3766

Page | 14
 9.6 0.01 0.01 0.16 9.5996 33.1146 0.5723
4.8 0.01 0.01 0.0226 23.8789 6.3982 0.1036
4.8 0.01 0.01 0.1845 7.4643 42.3434 0.7685
4.8 0.01 0.01 0.0557 17.2319 13.7532 0.2197
4.8 0.01 0.01 0.0982 13.3845 21.4178 0.5364
4.8 0.01 0.01 0.0964 12.7041 23.1631 0.5562
4.8 0.01 0.01 0.0706 15.5438 16.7035 0.366
4.8 0.01 0.01 0.0687 16.6351 14.7314 0.5191
4.8 0.01 0.01 0.0286 23.2529 6.8763 0.1626
4.8 0.01 0.01 0.0539 20.1341 9.8468 0.2535
128 4.8 0.01 0.01 0.0316 24.0534 6.2709 0.1003
4.8 0.01 0.01 0.0798 15.0647 17.6509 0.4761
4.8 0.01 0.01 0.1255 8.1654 39.0598 0.4647
4.8 0.01 0.01 0.0447 20.2804 9.6823 0.2416
4.8 0.01 0.01 0.0313 22.1798 7.7806 0.1502
4.8 0.01 0.01 0.0903 13.429 21.3083 0.4416
4.8 0.01 0.01 0.0661 17.8037 12.8769 0.36
4.8 0.01 0.01 0.0485 19.5318 10.5538 0.3047
4.8 0.01 0.01 0.0609 17.7767 12.9171 0.3052
2.4 0.01 0.01 0.0104 30.5997 2.9513 0.0805
2.4 0.01 0.01 0.0537 18.187 12.3211 0.4494
2.4 0.01 0.01 0.018 27.0248 4.4541 0.1435
2.4 0.01 0.01 0.0379 21.664 8.2566 0.2768
2.4 0.01 0.01 0.0471 18.919 11.3252 0.3744
2.4 0.01 0.01 0.0318 22.464 7.5301 0.2288
2.4 0.01 0.01 0.0351 22.475 7.5206 0.2901
2.4 0.01 0.01 0.0125 30.4439 3.0047 0.1081
2.4 0.01 0.01 0.0145 31.5098 2.6577 0.1039
256
2.4 0.01 0.01 0.0087 35.273 1.7232 0.0696
2.4 0.01 0.01 0.0409 20.877 9.0396 0.3222
2.4 0.01 0.01 0.0404 18.0062 12.5802 0.3179
2.4 0.01 0.01 0.0198 27.3261 4.3022 0.1567
2.4 0.01 0.01 0.0148 28.6925 3.676 0.1093
2.4 0.01 0.01 0.0308 22.7829 7.2586 0.2102
2.4 0.01 0.01 0.0296 24.7729 5.7724 0.2179
2.4 0.01 0.01 0.0194 27.5042 4.2149 0.196
2.4 0.01 0.01 0.0263 25.0585 5.5857 0.2083

Table 5 Performance of slp database for measurements of M =64, 128, and 256 for sine and cosine
dictionary.
M CR λ Rel. Tot RMSE SNR PRD MSE
9.6 0.01 0.01 0.0594 15.4902 16.8069 0.2251
9.6 0.01 0.01 0.1161 11.4906 26.6361 0.4038
9.6 0.01 0.01 0.1152 10.917 28.4543 0.4312
9.6 0.01 0.01 0.149 9.7605 32.507 0.5098
9.6 0.01 0.01 0.1612 8.2383 38.7334 0.4492
9.6 0.01 0.01 0.0826 14.1797 19.5441 0.2646
9.6 0.01 0.01 0.1031 13.1192 22.0821 0.314
9.6 0.01 0.01 0.1301 10.1022 31.253 0.4367
9.6 0.01 0.01 0.1121 13.7703 20.4874 0.2841
64
9.6 0.01 0.01 0.1232 12.2346 24.4495 0.3839
9.6 0.01 0.01 0.072 15.9503 15.9399 0.3101
9.6 0.01 0.01 0.1457 6.8753 45.3141 0.4711
9.6 0.01 0.01 0.0575 18.089 12.4609 0.195
9.6 0.01 0.01 0.1203 10.498 29.8606 0.3505
9.6 0.01 0.01 0.1106 11.6677 26.0986 0.3746
9.6 0.01 0.01 0.1599 10.1368 31.1286 0.4498
9.6 0.01 0.01 0.1307 10.9116 28.472 0.5603
9.6 0.01 0.01 0.1005 13.4332 21.298 0.3779
4.8 0.01 0.01 0.0119 29.4851 3.3554 0.074
4.8 0.01 0.01 0.0247 24.9402 5.6623 0.2501
128
4.8 0.01 0.01 0.0209 25.7514 5.1574 0.0566

Page | 15
 4.8 0.01 0.01 0.0235 25.811 5.1221 0.1629
4.8 0.01 0.01 0.0218 25.6152 5.2389 0.1728
4.8 0.01 0.01 0.0233 25.1897 5.5019 0.1709
4.8 0.01 0.01 0.0345 22.6114 7.4034 0.1833
4.8 0.01 0.01 0.011 31.5627 2.6416 0.0581
4.8 0.01 0.01 0.0389 22.9644 7.1086 0.1609
4.8 0.01 0.01 0.0324 23.8476 6.4213 0.1
4.8 0.01 0.01 0.0199 27.122 4.4045 0.1947
4.8 0.01 0.01 0.0562 15.1432 17.4921 0.3063
4.8 0.01 0.01 0.0174 28.4543 3.7782 0.0751
4.8 0.01 0.01 0.0257 23.9015 6.3815 0.125
4.8 0.01 0.01 0.0402 20.468 9.4754 0.194
4.8 0.01 0.01 0.0239 26.6345 4.6588 0.1361
4.8 0.01 0.01 0.0412 20.93 8.9847 0.3451
4.8 0.01 0.01 0.0183 28.2233 3.88 0.1209
2.4 0.01 0.01 0.015 27.4357 4.2483 0.0675
2.4 0.01 0.01 0.0119 31.2529 2.7375 0.1469
2.4 0.01 0.01 0.0113 32.1264 2.4756 0.0938
2.4 0.01 0.01 0.013 30.1207 3.1186 0.1042
2.4 0.01 0.01 0.013 30.1207 3.1186 0.1042
2.4 0.01 0.01 0.0133 30.0273 3.1524 0.1147
2.4 0.01 0.01 0.012 31.8043 2.5691 0.107
2.4 0.01 0.01 0.0068 35.7211 1.6366 0.0388
2.4 0.01 0.01 0.0095 35.2212 1.7336 0.0837
256
2.4 0.01 0.01 0.0064 37.8911 1.2748 0.0196
2.4 0.01 0.01 0.0202 26.9845 4.4748 0.1378
2.4 0.01 0.01 0.0134 27.6235 4.1574 0.1492
2.4 0.01 0.01 0.0099 33.3673 2.146 0.0519
2.4 0.01 0.01 0.0121 30.4237 3.0117 0.103
2.4 0.01 0.01 0.0149 29.0756 3.5174 0.1495
2.4 0.01 0.01 0.0107 33.6366 2.0805 0.0835
2.4 0.01 0.01 0.0156 29.3959 3.39 0.1937
2.4 0.01 0.01 0.0086 34.7734 1.8253 0.0619

Table 6 Performance of slp database for measurements of M =64, 128, and 256 for sine, cosine and identity
dictionary.
M CR λ Rel. Tot RMSE SNR PRD MSE
9.6 0.01 0.01 0.1245 9.0685 35.2025 0.4853
9.6 0.01 0.01 0.1442 9.6051 33.0936 0.4224
9.6 0.01 0.01 0.1288 9.9427 31.8323 0.4864
9.6 0.01 0.01 0.1783 8.2015 38.8979 0.5887
9.6 0.01 0.01 0.1696 7.7959 40.7572 0.4866
9.6 0.01 0.01 0.1355 9.8806 32.0604 0.383
9.6 0.01 0.01 0.1218 11.6704 26.0905 0.3112
9.6 0.01 0.01 0.148 8.9823 35.5539 0.4776
9.6 0.01 0.01 0.1459 11.4763 26.6799 0.3867
64
9.6 0.01 0.01 0.164 9.7484 32.552 0.4859
9.6 0.01 0.01 0.0944 13.6049 20.8811 0.3269
9.6 0.01 0.01 0.1497 6.64 46.5588 0.5762
9.6 0.01 0.01 0.0787 15.3641 17.0528 0.2859
9.6 0.01 0.01 0.1478 8.7065 36.7008 0.401
9.6 0.01 0.01 0.1858 7.1632 43.8368 0.5913
9.6 0.01 0.01 0.1914 8.5727 37.2705 0.5496
9.6 0.01 0.01 0.1419 10.2008 30.9001 0.57
9.6 0.01 0.01 0.1357 10.8196 28.7752 0.5307
4.8 0.01 0.01 0.0223 24.0113 6.3013 0.1067
4.8 0.01 0.01 0.0402 20.6987 9.2271 0.3074
4.8 0.01 0.01 0.0403 20.0295 9.9661 0.1252
128 4.8 0.01 0.01 0.0352 22.291 7.6815 0.1787
4.8 0.01 0.01 0.035 21.4926 8.4211 0.1996
4.8 0.01 0.01 0.0316 22.5219 7.4801 0.2174

Page | 16
 4.8 0.01 0.01 0.0386 21.6484 8.2715 0.185
4.8 0.01 0.01 0.0281 23.4274 6.7395 0.0955
4.8 0.01 0.01 0.042 22.286 7.686 0.1931
4.8 0.01 0.01 0.0544 19.3354 10.7952 0.1828
4.8 0.01 0.01 0.0286 23.9609 6.338 0.2339
4.8 0.01 0.01 0.0953 10.5618 29.6423 0.3687
4.8 0.01 0.01 0.0277 24.4397 5.9981 0.0972
4.8 0.01 0.01 0.0455 18.9322 11.3081 0.1807
4.8 0.01 0.01 0.052 18.2149 12.2816 0.2679
4.8 0.01 0.01 0.0317 24.1995 6.1663 0.1552
4.8 0.01 0.01 0.0489 19.4432 10.662 0.4061
4.8 0.01 0.01 0.0282 24.477 5.9724 0.1309
2.4 0.01 0.01 0.0177 25.9972 5.0135 0.1018
2.4 0.01 0.01 0.0289 23.5765 6.6248 0.2741
2.4 0.01 0.01 0.0187 26.7056 4.6208 0.089
2.4 0.01 0.01 0.0271 24.5799 5.9021 0.1836
2.4 0.01 0.01 0.0286 23.2501 6.8785 0.2151
2.4 0.01 0.01 0.0261 24.1904 6.1727 0.2048
2.4 0.01 0.01 0.0307 23.6329 6.5819 0.1508
2.4 0.01 0.01 0.0425 19.8196 10.2099 0.1765
2.4 0.01 0.01 0.0245 26.9625 4.4862 0.1679
256
2.4 0.01 0.01 0.0299 24.5429 5.9273 0.0945
2.4 0.01 0.01 0.0274 24.3487 6.0613 0.2281
2.4 0.01 0.01 0.0299 20.6153 9.3161 0.3229
2.4 0.01 0.01 0.0316 23.275 6.8588 0.1097
2.4 0.01 0.01 0.0236 24.6507 5.8542 0.1586
2.4 0.01 0.01 0.0265 24.0708 6.2584 0.2458
2.4 0.01 0.01 0.0266 25.7258 5.1726 0.1442
2.4 0.01 0.01 0.0718 16.119 15.6333 0.3408
2.4 0.01 0.01 0.0244 25.7358 5.1666 0.1276

Table 7 Performance of ourdatabase database for measurements of M =64, 128, and 256 for cosine
dictionary.
M CR λ Rel. Tot RMSE SNR PRD MSE
9.6 0.01 0.01 0.3731 1.5738 83.4279 0.947
9.6 0.01 0.01 0.3444 0.9106 90.0467 1.117
9.6 0.01 0.01 0.3146 2.3954 75.8978 0.8477
9.6 0.01 0.01 0.1205 11.9652 25.2197 0.3642
9.6 0.01 0.01 0.1003 11.7294 25.9136 0.3893
9.6 0.01 0.01 0.2562 5.5242 52.9408 0.8901
9.6 0.01 0.01 0.2761 5.4389 53.4634 0.9699
9.6 0.01 0.01 0.1884 8.0808 39.4421 0.5581
9.6 0.01 0.01 0.1725 7.9434 40.0711 0.5355
9.6 0.01 0.01 0.2425 3.8442 64.2376 0.9183
64 9.6 0.01 0.01 0.2796 5.0274 56.0572 0.778
9.6 0.01 0.01 0.2176 7.2738 43.2824 0.6464
9.6 0.01 0.01 0.1206 10.171 31.0064 0.4782
9.6 0.01 0.01 0.2855 5.7354 51.6687 1.0593
9.6 0.01 0.01 0.2945 4.1914 61.7205 0.9322
9.6 0.01 0.01 0.2538 6.5104 47.2588 0.7088
9.6 0.01 0.01 0.2588 6.4512 47.5816 1.0478
9.6 0.01 0.01 0.1657 9.0625 35.2268 0.4049
9.6 0.01 0.01 0.2249 6.1853 49.0609 0.817
9.6 0.01 0.01 0.3437 2.4843 75.1254 1.0757
9.6 0.01 0.01 0.1234 10.8513 28.6704 0.3677
9.6 0.01 0.01 0.1704 8.4322 37.8781 0.6796
4.8 0.01 0.01 0.0903 13.8957 20.1936 0.4545
128 4.8 0.01 0.01 0.117 10.2887 30.5889 0.4429

Page | 17
 4.8 0.01 0.01 0.1559 8.4932 37.613 0.4551
4.8 0.01 0.01 0.0149 30.1469 3.1092 0.0664
4.8 0.01 0.01 0.0123 29.9638 3.1755 0.061
4.8 0.01 0.01 0.0604 18.0728 12.4841 0.2691
4.8 0.01 0.01 0.0639 18.1504 12.3732 0.2307
4.8 0.01 0.01 0.0845 15.0397 17.7016 0.4194
4.8 0.01 0.01 0.0375 21.1975 8.7121 0.1241
4.8 0.01 0.01 0.1413 8.5366 37.4259 0.6892
4.8 0.01 0.01 0.0371 22.567 7.4413 0.1386
4.8 0.01 0.01 0.0696 17.1798 13.836 0.4005
4.8 0.01 0.01 0.0436 19.0127 11.2038 0.2516
4.8 0.01 0.01 0.1172 13.466 21.2178 0.5433
4.8 0.01 0.01 0.0625 17.651 13.1053 0.3353
4.8 0.01 0.01 0.0724 17.4019 13.4866 0.3721
4.8 0.01 0.01 0.1385 11.8807 25.4662 0.8509
4.8 0.01 0.01 0.0473 19.9435 10.0652 0.2067
4.8 0.01 0.01 0.1002 13.2041 21.8673 0.5387
4.8 0.01 0.01 0.0384 21.5261 8.3887 0.2323
4.8 0.01 0.01 0.0865 14.3239 19.2222 0.4729
4.8 0.01 0.01 0.0547 17.9181 12.7086 0.2667
2.4 0.01 0.01 0.0413 20.7007 9.225 0.2603
2.4 0.01 0.01 0.0311 21.7956 8.1325 0.2219
2.4 0.01 0.01 0.0373 20.9236 8.9913 0.279
2.4 0.01 0.01 0.0053 39.0187 1.1196 0.0416
2.4 0.01 0.01 0.0052 37.373 1.3532 0.0561
2.4 0.01 0.01 0.0272 25.015 5.6137 0.1846
2.4 0.01 0.01 0.023 27.0217 4.4557 0.1549
2.4 0.01 0.01 0.041 21.3253 8.5849 0.2802
2.4 0.01 0.01 0.0117 31.3356 2.7116 0.0727
2.4 0.01 0.01 0.0649 15.2911 17.1968 0.4498
2.4 0.01 0.01 0.0117 32.5809 2.3494 0.0665
256
2.4 0.01 0.01 0.0367 22.7344 7.2993 0.2858
2.4 0.01 0.01 0.0245 24.0216 6.2939 0.1687
2.4 0.01 0.01 0.0461 21.5807 8.3361 0.3658
2.4 0.01 0.01 0.0281 24.5953 5.8916 0.2004
2.4 0.01 0.01 0.0296 25.1767 5.5102 0.2426
2.4 0.01 0.01 0.0617 18.8989 11.3515 0.462
2.4 0.01 0.01 0.0241 25.8036 5.1265 0.1408
2.4 0.01 0.01 0.0415 20.8692 9.0477 0.3287
2.4 0.01 0.01 0.0195 27.4029 4.2644 0.1646
2.4 0.01 0.01 0.0449 20.0189 9.9783 0.3293
2.4 0.01 0.01 0.0165 28.3259 3.8345 0.1755

Table 8 Performance of ourdatabase database for measurements of M =64, 128, and 256 for sine and cosine
dictionary.

M CR λ Rel. Tot RMSE SNR PRD MSE

9.6 0.01 0.01 0.3213 2.8737 71.8314 0.9578
9.6 0.01 0.01 0.2288 4.4649 59.8073 0.5337
9.6 0.01 0.01 0.2713 3.6832 65.4394 0.6213
9.6 0.01 0.01 0.0378 22.0409 7.9059 0.109
9.6 0.01 0.01 0.0613 16.0095 15.8317 0.1965
9.6 0.01 0.01 0.0795 15.6837 16.4366 0.2968
9.6 0.01 0.01 0.0973 14.4943 18.8488 0.2608
64 9.6 0.01 0.01 0.0951 14.0214 19.9036 0.2774
9.6 0.01 0.01 0.1053 12.235 24.4482 0.3053
9.6 0.01 0.01 0.0772 13.7808 20.4626 0.3552
9.6 0.01 0.01 0.0826 15.6219 16.554 0.2029
9.6 0.01 0.01 0.0807 15.8938 16.044 0.2229
9.6 0.01 0.01 0.0629 15.8333 16.156 0.1834
9.6 0.01 0.01 0.1003 14.8252 18.1443 0.2902
9.6 0.01 0.01 0.0965 13.8868 20.2144 0.2907
9.6 0.01 0.01 0.1067 14.0397 19.8617 0.2245

Page | 18
 9.6 0.01 0.01 0.1429 11.6086 26.2766 0.5785
9.6 0.01 0.01 0.12 11.8635 25.5168 0.3802
9.6 0.01 0.01 0.1087 12.4987 23.7174 0.4304
9.6 0.01 0.01 0.1601 9.1216 34.9881 0.3801
9.6 0.01 0.01 0.0709 16.0494 15.759 0.3499
9.6 0.01 0.01 0.1267 10.6194 29.4461 0.3189
4.8 0.01 0.01 0.0568 17.9251 12.6983 0.2092
4.8 0.01 0.01 0.0692 14.8489 18.0948 0.2042
4.8 0.01 0.01 0.0778 14.5266 18.7789 0.3655
4.8 0.01 0.01 0.0099 33.6399 2.0797 0.0368
4.8 0.01 0.01 0.0097 32.0423 2.4997 0.0686
4.8 0.01 0.01 0.0303 24.0611 6.2653 0.1943
4.8 0.01 0.01 0.0295 24.8757 5.7045 0.26
4.8 0.01 0.01 0.02 27.562 4.187 0.1552
4.8 0.01 0.01 0.0199 26.6921 4.628 0.0575
4.8 0.01 0.01 0.027 22.9138 7.1501 0.2286
4.8 0.01 0.01 0.0257 25.7693 5.1468 0.0755
128 4.8 0.01 0.01 0.0213 27.4664 4.2333 0.1568
4.8 0.01 0.01 0.0149 28.3307 3.8323 0.1068
4.8 0.01 0.01 0.0278 25.9767 5.0253 0.2436
4.8 0.01 0.01 0.0218 26.8142 4.5634 0.1658
4.8 0.01 0.01 0.0528 20.2631 9.7016 0.3191
4.8 0.01 0.01 0.0307 23.7002 6.5312 0.1946
4.8 0.01 0.01 0.0394 21.3084 8.6016 0.2816
4.8 0.01 0.01 0.017 28.601 3.7149 0.133
4.8 0.01 0.01 0.0232 25.7435 5.1621 0.1897
4.8 0.01 0.01 0.0459 19.4426 10.6627 0.3095
2.4 0.01 0.01 0.0263 24.6019 5.8872 0.1387
2.4 0.01 0.01 0.0227 24.5464 5.9249 0.1232
2.4 0.01 0.01 0.0216 25.6469 5.2198 0.1721
2.4 0.01 0.01 0.0074 36.2404 1.5416 0.0281
2.4 0.01 0.01 0.0084 33.29 2.1652 0.0582
2.4 0.01 0.01 0.0113 32.6415 2.333 0.1105
2.4 0.01 0.01 0.0128 32.0913 2.4856 0.1484
2.4 0.01 0.01 0.0102 33.4472 2.1264 0.1007
2.4 0.01 0.01 0.0051 38.4946 1.1892 0.0211
2.4 0.01 0.01 0.0118 30.1259 3.1168 0.1378
2.4 0.01 0.01 0.0084 35.5115 1.6766 0.0657
2.4 0.01 0.01 0.0108 33.3361 2.1538 0.0991
2.4 0.01 0.01 0.0128 29.6488 3.2928 0.0707
256 2.4 0.01 0.01 0.0159 30.8054 2.8822 0.1843
2.4 0.01 0.01 0.0096 33.8899 2.0207 0.0997
2.4 0.01 0.01 0.0086 35.908 1.6018 0.0556
2.4 0.01 0.01 0.0349 23.8536 6.4168 0.2036
2.4 0.01 0.01 0.0118 32.0047 2.5105 0.1157
2.4 0.01 0.01 0.0192 27.5467 4.1944 0.1563
2.4 0.01 0.01 0.0101 33.1432 2.2021 0.0849
2.4 0.01 0.01 0.0106 32.5791 2.3499 0.1284
2.4 0.01 0.01 0.013 30.3655 3.032 0.1671

Table 9 Performance of ourdatabase database for measurements of M =64, 128, and 256 for sine, cosine and
identity dictionary.
M CR λ Rel. Tot RMSE SNR PRD MSE
9.6 0.01 0.01 0.3327 2.5695 74.3919 0.9531
9.6 0.01 0.01 0.2857 2.5334 74.7019 0.6414
9.6 0.01 0.01 0.3204 2.2377 77.2887 0.8346
9.6 0.01 0.01 0.0565 18.542 11.8276 0.158
9.6 0.01 0.01 0.071 14.7354 18.3328 0.2492
9.6 0.01 0.01 0.1083 12.9997 22.3881 0.3135
9.6 0.01 0.01 0.1073 13.6494 20.7745 0.2788
9.6 0.01 0.01 0.1465 10.2631 30.6792 0.4415
64
9.6 0.01 0.01 0.1112 11.7579 25.8289 0.3033
9.6 0.01 0.01 0.1021 11.3597 27.0407 0.3636
9.6 0.01 0.01 0.1222 12.2156 24.5029 0.2596
9.6 0.01 0.01 0.101 13.9444 20.0807 0.3163
9.6 0.01 0.01 0.0801 13.7301 20.5824 0.2318
9.6 0.01 0.01 0.122 13.1242 22.0693 0.3455
9.6 0.01 0.01 0.1131 12.5076 23.693 0.3446
9.6 0.01 0.01 0.1777 9.6063 33.0892 0.3514

Page | 19
 9.6 0.01 0.01 0.1714 10.0304 31.5122 0.6714
9.6 0.01 0.01 0.1214 11.7658 25.8054 0.3777
9.6 0.01 0.01 0.1551 9.4107 33.8426 0.4391
9.6 0.01 0.01 0.2436 5.4754 53.2393 0.5655
9.6 0.01 0.01 0.0787 15.138 17.5025 0.3703
9.6 0.01 0.01 0.1241 10.8011 28.8366 0.473
4.8 0.01 0.01 0.0954 13.4237 21.3214 0.2565
4.8 0.01 0.01 0.1246 9.7431 32.5719 0.3114
4.8 0.01 0.01 0.1193 10.8182 28.78 0.426
4.8 0.01 0.01 0.0206 27.3088 4.3108 0.0532
4.8 0.01 0.01 0.0184 26.4652 4.7505 0.0974
4.8 0.01 0.01 0.0313 23.7953 6.46 0.1991
4.8 0.01 0.01 0.0379 22.6787 7.3463 0.3073
4.8 0.01 0.01 0.0312 23.6857 6.5421 0.1932
4.8 0.01 0.01 0.0335 22.1692 7.7901 0.0927
4.8 0.01 0.01 0.0368 20.2238 9.7456 0.2812
4.8 0.01 0.01 0.0319 23.8764 6.4 0.0856
128
4.8 0.01 0.01 0.0302 24.4369 6.0001 0.1941
4.8 0.01 0.01 0.0229 24.6095 5.882 0.1406
4.8 0.01 0.01 0.039 23.0292 7.0557 0.332
4.8 0.01 0.01 0.0336 23.0441 7.0436 0.1835
4.8 0.01 0.01 0.0355 23.5887 6.6155 0.0911
4.8 0.01 0.01 0.038 23.1184 6.9836 0.305
4.8 0.01 0.01 0.0365 22.1971 7.765 0.2396
4.8 0.01 0.01 0.0561 18.2501 12.2319 0.3123
4.8 0.01 0.01 0.0304 23.547 6.6473 0.1721
4.8 0.01 0.01 0.0328 22.7513 7.2851 0.2624
4.8 0.01 0.01 0.0603 17.0687 14.0141 0.3704
2.4 0.01 0.01 0.0603 17.4074 13.4781 0.2243
2.4 0.01 0.01 0.0597 16.1392 15.597 0.2281
2.4 0.01 0.01 0.0567 17.2801 13.6771 0.2808
2.4 0.01 0.01 0.0184 28.2823 3.8538 0.0519
2.4 0.01 0.01 0.0156 27.9169 4.0193 0.0917
2.4 0.01 0.01 0.0231 26.4214 4.7745 0.1822
2.4 0.01 0.01 0.0295 24.8673 5.71 0.282
2.4 0.01 0.01 0.0357 22.5364 7.4676 0.197
2.4 0.01 0.01 0.0175 27.8082 4.0699 0.0473
2.4 0.01 0.01 0.0267 23.0211 7.0623 0.2659
2.4 0.01 0.01 0.0257 25.7454 5.161 0.0786
2.4 0.01 0.01 0.0227 26.9234 4.5064 0.1863
2.4 0.01 0.01 0.0161 27.6618 4.1392 0.1235
2.4 0.01 0.01 0.034 24.2171 6.1538 0.3327
2.4 0.01 0.01 0.0314 23.6271 6.5863 0.1771
2.4 0.01 0.01 0.0245 26.8252 4.5577 0.0655
2.4 0.01 0.01 0.028 25.7688 5.1471 0.2915
2.4 0.01 0.01 0.0281 24.4796 5.9707 0.2346
2.4 0.01 0.01 0.0359 22.1177 7.8364 0.2806
2.4 0.01 0.01 0.0325 22.961 7.1113 0.1556
2.4 0.01 0.01 0.0286 23.9202 6.3678 0.2619
256 2.4 0.01 0.01 0.0268 24.1023 6.2357 0.3127

The waveforms is given in figure 6(a), (b) and (c) below shows measurements for
reconstruction of original signal, error signal and the reconstructed signal also. For example
We considered a signal taken from the MIMIC database with length N=512 and M=256. The
figure 6(a) ), (b) and (c) shows the original signal, measurements, reconstructed signal and
error signal for cosine dictionary, sine cosine dictionary and sine cosine identity dictionary
respectively.

Page | 20
 𝒂 𝒃

Fig 6(a),(b)and(c): Illustrates the CS results of the proposed methods for signal taken from the
MIMIC database with M=256

Page | 21
DCT based PPG compression algorithm is implemented with the help of Matlab and
compared with the CS based data reduction algorithm. The quality of the waveform is also
analysed while comparing the performance of both algorithms. The DCT based PPG
compression algorithm shows better quality of waveform along with the below 10 PRD
value for higher compression ratio.

The Tables from 10- 12 shows the 10 Comparison of performance of CS based and DCT
based PPG compression algorithm for MIMIC database, slp database, ourdatabase
respectively.

Table 10 Comparison of performance of CS based and DCT based PPG compression algorithm for MIMIC
database.
M CR λ Rel.T RMSE SNR PRD MAE QUALI CR RMSE SNR PRD MAE QUALI
ot TY TY
64 12 0.01 0.01 0.110 13.312 21.595 0.395 3 12 0.052 19.847 10.176 0.718 5
5 9 9 1 9 6 5
12 0.01 0.01 0.110 13.312 21.595 0.395 3 12 0.052 19.847 10.176 0.718 5
5 9 9 1 9 6 5
12 0.01 0.01 0.095 13.286 21.660 0.316 3 12 0.023 25.465 5.3302 0.323 5
7 3 1 4 2 9
12 0.01 0.01 0.049 18.234 12.253 0.250 4 12 0.021 25.572 5.265 0.235 4
1 5 9 3 1 1 6
12 0.01 0.01 0.091 12.339 24.155 0.550 3 12 0.016 27.374 4.2785 0.342 5
5 6 7 3 2 1 5
12 0.01 0.01 0.090 13.892 20.200 0.377 3 12 0.036 21.879 8.0542 0.472 4
7 6 9 9 2 6 1
12 0.01 0.01 0.129 11.519 26.546 0.476 3 12 0.022 26.702 4.6223 0.337 5
6 7 8 4 6 9 4
12 0.01 0.01 0.056 17.869 12.780 0.277 4 12 0.023 25.371 5.3881 0.291 5
6 5 5 9 4 9
12 0.01 0.01 0.059 16.979 14.159 0.265 4 12 0.021 25.874 5.0848 0.287 4
4 3 2 2 3 5 5
12 0.01 0.01 0.137 5.7068 51.839 1.108 3 12 0.067 11.975 25.189 1.045 4
9 7 8 6 6 6
12 0.01 0.01 0.056 19.397 10.718 0.354 3 12 0.043 21.708 8.2148 0.621 5
1 1 8 2
12 0.01 0.01 0.060 17.660 13.091 0.467 4 12 0.015 29.282 3.4345 0.319 4
6 4 2 1 9 8 9
12 0.01 0.01 0.043 20.534 9.4033 0.413 4 12 0.02 27.209 4.3605 0.382 4
2 4 8 3 6
12 0.01 0.01 0.057 17.941 12.674 0.332 3 12 0.029 23.768 6.48 0.389 5
3 6 1 9 3 5 7
12 0.01 0.01 0.073 16.658 14.692 0.485 4 12 0.038 22.222 7.7428 0.556 4
2 1 5 4 6 1 9
12 0.01 0.01 0.059 18.107 12.434 0.350 3 12 0.019 27.854 4.0483 0.219 4
2 2 8 4 3 5
12 0.01 0.01 0.058 18.390 12.035 0.399 4 12 0.019 27.829 4.06 0.335 5
3 7 6 6 7 5 5
12 0.01 0.01 0.071 15.753 16.305 0.443 3 12 0.026 24.458 5.9853 0.315 5
5 2 7 5 3 3 2
12 0.01 0.01 0.067 16.960 14.190 0.359 4 12 0.023 26.160 4.9203 0.351 5
4 1 5 2 4 1 2
12 0.01 0.01 0.063 17.903 12.729 0.531 3 12 0.043 21.122 8.7878 0.557 4
5 7 7 7 8 4
12 0.01 0.01 0.079 15.717 16.372 0.775 3 12 0.032 23.351 6.799 0.410 4
5 8 6 8 2 2
128 6 0.01 0.01 0.034 23.401 6.7593 0.410 4 6 0.035 23.099 6.9989 0.487 5
6 9 8 8 5 3
6 0.01 0.01 0.034 23.401 6.7593 0.410 4 6 0.035 23.099 6.9989 0.487 5
6 9 8 8 5 3
6 0.01 0.01 0.034 22.131 7.8237 0.428 4 6 0.022 25.658 5.213 0.290 4
3 7 9 3 6

Page | 22
 6 0.01 0.01 0.020 25.838 5.106 0.266 4 6 0.018 26.528 4.7161 0.210 5
4 4 1 9 3 7
6 0.01 0.01 0.033 21.004 8.9076 0.378 4 6 0.016 27.374 4.2785 0.342 4
7 8 4 2 1 5
6 0.01 0.01 0.035 22.076 7.8733 0.397 4 6 0.026 24.492 5.9621 0.332 4
4 9 8 8 1 9
6 0.01 0.01 0.027 24.953 5.6535 0.298 4 6 0.019 28.198 3.8913 0.332 4
6 6 2 1
6 0.01 0.01 0.031 23.071 7.0211 0.326 4 6 0.020 26.860 4.539 0.201 4
1 9 8 1 8
6 0.01 0.01 0.031 22.449 7.5429 0.250 4 6 0.016 27.933 4.0118 0.285 5
7 3 9 8 2 8
6 0.01 0.01 0.078 10.649 29.344 1.173 4 6 0.064 12.266 24.358 0.855 4
6 5 8 9 6 4
6 0.01 0.01 0.040 22.265 7.7046 0.326 4 6 0.028 25.240 5.47 0.432 5
3 2 6 3 2
6 0.01 0.01 0.030 23.556 6.6398 0.347 4 6 0.016 29.240 3.4513 0.321 5
7 9 5 3 9
6 0.01 0.01 0.028 24.059 6.2664 0.350 4 6 0.017 28.378 3.8113 0.348 5
8 6 7 5 5
6 0.01 0.01 0.025 25.035 5.6007 0.317 4 6 0.022 26.005 5.009 0.272 5
3 1 8 7 8
6 0.01 0.01 0.029 24.599 5.8887 0.294 4 6 0.026 25.435 5.3487 0.392 5
4 6 7 1
6 0.01 0.01 0.023 26.086 4.9622 0.297 4 6 0.017 28.687 3.6783 0.204 5
6 5 3 5 2 7
6 0.01 0.01 0.027 25.085 5.5684 0.309 4 6 0.019 27.829 4.06 0.335 5
3 4 7 5 5
6 0.01 0.01 0.031 22.818 7.229 0.389 4 6 0.023 25.331 5.4128 0.282 5
7 4 5 7 5 4
6 0.01 0.01 0.03 23.999 6.3102 0.447 4 6 0.022 26.390 4.7913 0.315 5
2 6 8 9 8
6 0.01 0.01 0.035 22.948 7.1217 0.515 4 6 0.031 24.064 6.2631 0.391 5
5 4 5 2 2
6 0.01 0.01 0.052 19.258 10.891 0.75 4 6 0.027 24.882 5.7003 0.361 5
6 7 5 1 3
256 3 0.01 0.01 0.025 26.147 4.9273 0.355 4 3 0.026 25.718 5.1769 0.499 5
2 7 8 5 6 3
3 0.01 0.01 0.025 26.147 4.9273 0.355 4 3 0.026 25.718 5.1769 0.499 5
2 7 8 5 6 3
3 0.01 0.01 0.023 25.595 5.251 0.357 5 3 0.021 26.132 4.9362 0.427 5
2 5 6 1 3
3 0.01 0.01 0.016 27.826 4.0613 0.268 4 3 0.018 26.811 4.565 0.319 4
3 7 4 3 1 7
3 0.01 0.01 0.016 27.488 4.2223 0.213 5 3 0.016 27.374 4.2785 0.342 5
9 2 1 5
3 0.01 0.01 0.025 24.938 5.6632 0.371 4 3 0.020 26.779 4.5817 0.384 5
4 7 9 6 4 5
3 0.01 0.01 0.015 29.979 3.1699 0.175 5 3 0.019 27.969 3.9953 0.346 5
5 2 7 5 4
3 0.01 0.01 0.016 28.436 3.7861 0.250 5 3 0.018 27.548 4.1935 0.333 5
8 3 7 6 4 8
3 0.01 0.01 0.014 29.258 3.4443 0.138 5 3 0.019 26.715 4.6157 0.390 5
5 9 4 2 8
3 0.01 0.01 0.071 11.435 26.806 1.198 4 3 0.054 13.769 20.488 0.957 5
3 2 5 5 5 8 4 8
3 0.01 0.01 0.018 29.016 3.5413 0.178 4 3 0.023 27.139 4.3955 0.331 5
5 7 3 8 1
3 0.01 0.01 0.016 29.174 3.4776 0.202 5 3 0.016 29.240 3.4513 0.321 5
1 5 4 3 9
3 0.01 0.01 0.017 28.231 3.8764 0.215 4 3 0.017 28.378 3.8113 0.348 5
8 5 7 5 5
3 0.01 0.01 0.018 27.574 4.1811 0.272 4 3 0.020 26.899 4.5188 0.340 5
9 3 6 4 6 9
3 0.01 0.01 0.020 27.902 4.026 0.189 4 3 0.022 27.05 4.4412 0.422 5
1 5 8 1 9
3 0.01 0.01 0.016 29.100 3.5072 0.251 5 3 0.019 27.952 4.0029 0.367 5
7 9 2 1 5 8
3 0.01 0.01 0.018 28.234 3.8752 0.206 5 3 0.025 25.451 5.3388 0.509 5

Page | 23
 8 1 2 9 2 6
3 0.01 0.01 0.023 25.504 5.3064 0.361 5 3 0.022 25.877 5.0834 0.413 5
3 7 3 7
3 0.01 0.01 0.026 25.249 5.4643 0.414 4 3 0.019 27.639 4.1497 0.386 5
3 7 7 8 7
3 0.01 0.01 0.035 23.022 7.0611 0.479 4 3 0.024 26.326 4.8269 0.440 4
2 6 1 1 5 7
3 0.01 0.01 0.035 22.740 7.2944 0.569 4 3 0.023 26.146 4.9278 0.462 5
2 3 5 8 9 1

Table 11 Comparison of performance of CS based and DCT based PPG compression algorithm for slp
database.
M CR λ Rel.T RMSE SNR PRD MAE QUAL CR RMSE SNR PRD MAE QUAL
ot ITY ITY
64 12 0.01 0.01 0.093 11.57 26.38 0.361 2 12 0.011 29.70 3.270 0.136 3
3 39 18 7 6 89 1 6
12 0.01 0.01 0.073 15.50 16.77 0.319 1 12 0.043 19.97 10.03 0.675 3
1 45 94 8 7 29 12 5
12 0.01 0.01 0.174 7.304 43.13 0.397 2 12 0.016 27.91 4.022 0.234 3
6 3 08 9 3 06 3
12 0.01 0.01 0.127 11.10 27.83 0.393 2 12 0.030 23.49 6.688 0.454 4
6 79 59 8 7 32 7 4
12 0.01 0.01 0.138 9.552 33.29 0.335 3 12 0.037 20.86 9.053 0.583 4
6 65 4 7 41
12 0.01 0.01 0.071 15.42 16.93 0.272 3 12 0.024 24.62 5.870 0.354 5
6 34 67 4 8 61 8 7
12 0.01 0.01 0.130 11.06 27.97 0.340 3 12 0.030 23.69 6.536 0.458 5
6 45 52 1 5 29 6 5
12 0.01 0.01 0.178 7.368 42.81 0.471 3 12 0.013 29.52 3.340 0.165 4
3 5 28 4 9 31 7 5
12 0.01 0.01 0.213 8.158 39.08 0.524 2 12 0.015 30.76 2.895 0.136 4
8 9 89 9 8 62 3 4
12 0.01 0.01 0.098 14.20 19.48 0.258 3 12 0.011 32.56 2.352 0.111 5
2 56 6 5 9 82 8 8
12 0.01 0.01 0.069 16.23 15.41 0.262 3 12 0.033 22.70 7.327 0.505 5
7 95 78 2 1 13 1 5
12 0.01 0.01 0.148 6.682 46.33 0.517 3 12 0.031 20.13 9.842 0.486 4
9 1 37 9 6 8 4 2
12 0.01 0.01 0.149 8.597 37.16 0.460 3 12 0.015 28.46 3.771 0.189 4
7 5 43 3 2 99 4 7
12 0.01 0.01 0.160 8.420 37.93 0.542 3 12 0.023 24.97 5.642 0.322 5
7 1 11 9 9 07 4 1
12 0.01 0.01 0.067 17.58 13.20 0.232 3 12 0.023 26.80 4.570 0.334 5
8 44 62 6 5 03 7 3
12 0.01 0.01 0.147 9.856 32.14 0.434 3 12 0.018 28.12 3.923 0.351 5
6 6 9 1 57 9 9
12 0.01 0.01 0.056 18.36 12.07 0.152 3 12 0.023 26.22 4.882 0.331 5
9 5 12 4 81 8
128 6 0.01 0.01 0.014 27.73 4.104 0.101 3 6 0.010 30.33 3.044 0.144 4
5 38 9 2 8 1 9
6 0.01 0.01 0.022 25.59 5.248 0.240 2 6 0.027 24.12 6.219 0.466 4
9 86 9 8 1 44 8 9
6 0.01 0.01 0.014 29.13 3.495 0.064 3 6 0.013 29.67 3.281 0.170 4
1 06 2 2 3 91 3 9
6 0.01 0.01 0.024 25.59 5.248 0.176 3 6 0.021 26.67 4.635 0.316 5
1 93 5 3 78 6
6 0.01 0.01 0.024 24.64 5.860 0.191 4 6 0.024 24.76 5.776 0.410 4
4 13 5 69 4 5
6 0.01 0.01 0.022 25.50 5.304 0.172 4 6 0.017 27.44 4.243 0.242 4
4 79 1 9 51 7
6 0.01 0.01 0.018 28.12 3.923 0.134 4 6 0.020 26.99 4.470 0.315 5
3 67 4 9 24 7 3

Page | 24
 6 0.01 0.01 0.010 31.79 2.570 0.060 4 6 0.013 30.05 3.142 0.165 5
7 82 9 1 1 5 3 8
6 0.01 0.01 0.022 27.92 4.017 0.152 4 6 0.015 30.95 2.832 0.180 5
01 9 3 5 63 6 3
6 0.01 0.01 0.032 23.94 6.347 0.105 4 6 0.014 30.96 2.829 0.176 4
86 3 51 7 4
6 0.01 0.01 0.016 28.95 3.566 0.188 4 6 0.021 26.32 4.827 0.350 4
1 62 1 5 8 57 4 1
6 0.01 0.01 0.032 19.80 10.22 0.284 4 6 0.019 24.31 6.086 0.336 4
9 6 59 4 6 29 3 4
6 0.01 0.01 0.028 23.09 7.002 0.140 4 6 0.014 29.04 3.528 0.224 4
2 55 1 9 2 85 4 7
6 0.01 0.01 0.028 23.45 6.720 0.247 4 6 0.017 27.61 4.162 0.269 5
5 22 3 3 6 23 8 8
6 0.01 0.01 0.023 26.88 4.523 0.128 4 6 0.018 28.90 3.587 0.231 4
2 97 9 6 4 32 9 4
6 0.01 0.01 0.052 18.83 11.43 0.359 3 6 0.016 28.74 3.655 0.333 4
5 9 01 4 8 23 2
6 0.01 0.01 0.015 29.46 3.363 0.106 4 6 0.018 28.18 3.899 0.238 5
9 49 2 9 4 1 2
256 3 0.01 0.01 0.008 32.47 2.377 0.062 4 3 0.012 28.90 3.588 0.195 5
4 81 4 1 7 15 6 7
3 0.01 0.01 0.013 30.03 3.151 0.154 3 3 0.017 27.71 4.115 0.235 4
7 06 2 6 9 2 3 9
3 0.01 0.01 0.007 34.70 1.840 0.044 4 3 0.013 29.33 3.413 0.187 5
4 21 3 2 8 53 8 8
3 0.01 0.01 0.012 31.54 2.647 0.099 4 3 0.018 28.10 3.935 0.313 5
1 46 1 6 1 1 4
3 0.01 0.01 0.017 27.69 4.122 0.124 4 3 0.017 27.49 4.219 0.228 4
2 66 6 6 44 7 6
3 0.01 0.01 0.033 22.15 7.806 0.117 4 3 0.016 28.01 3.974 0.304 5
07 6 4 8 44 5 6
3 0.01 0.01 0.012 31.49 2.662 0.094 4 3 0.017 28.33 3.829 0.274 5
4 51 2 3 9 77 2 2
3 0.01 0.01 0.007 35.29 1.718 0.052 4 3 0.015 28.35 3.822 0.227 4
2 48 9 7 9 35 3 6
3 0.01 0.01 0.009 34.93 1.791 0.065 4 3 0.019 29.14 3.489 0.256 4
8 51 6 3 1 54 3 8
3 0.01 0.01 0.023 26.57 4.689 0.065 4 3 0.020 27.79 4.073 0.288 4
6 69 8 7 5 98 9 4
3 0.01 0.01 0.015 29.15 3.484 0.140 4 3 0.018 27.89 4.030 0.236 4
7 68 6 1 2 24 7 6
3 0.01 0.01 0.014 26.73 4.605 0.164 4 3 0.013 27.67 4.134 0.175 5
8 52 1 1 3 07 9 4
3 0.01 0.01 0.014 28.73 3.659 0.077 4 3 0.018 26.84 4.546 0.331 5
7 08 8 8 3 71 1 2
3 0.01 0.01 0.012 30.52 2.978 0.132 4 3 0.017 27.63 4.153 0.333 4
6 05 3 2 6 17 5 1
3 0.01 0.01 0.010 33.74 2.054 0.075 5 3 0.016 29.94 3.180 0.261 5
6 61 5 4 3 99 6 9
3 0.01 0.01 0.014 29.75 3.251 0.173 4 3 0.016 28.74 3.655 0.333 4
9 74 9 6 8 23 2
3 0.01 0.01 0.008 34.72 1.835 0.076 5 3 0.018 28.28 3.851 0.251 5
7 66 1 2 65 9

Table 12 Comparison of performance of CS based and DCT based PPG compression algorithm for
ourdatabase database

M CR λ Rel.T RMSE SNR PRD MAE QUAL CR RMSE SNR PRD MAE QUAL
ot ITY ITY
64 12 0.01 0.01 0.268 4.433 60.02 0.742 2 12 0.031 23.11 6.983 0.443 4
4 5 38 1 2 81 9 7

Page | 25
 12 0.01 0.01 0.233 4.288 61.03 0.617 2 12 0.022 24.61 5.879 0.292 5
5 2 63 9 5 3 6 2
12 0.01 0.01 0.297 2.896 71.64 0.806 3 12 0.031 22.29 7.681 0.458 5
8 1 5 8 06 9 2
12 0.01 0.01 0.058 18.20 12.30 0.141 4 12 0.011 32.43 2.389 0.120 5
8 11 11 4 38 5 2
12 0.01 0.01 0.094 12.29 24.27 0.328 3 12 0.007 34.43 1.897 0.123 5
75 32 3 3 46 9 4
12 0.01 0.01 0.135 11.07 27.93 0.353 3 12 0.019 27.83 4.059 0.270 5
2 6 82 5 6 02 7 9
12 0.01 0.01 0.075 16.71 14.59 0.374 3 12 0.020 28.02 3.968 0.266 5
4 72 28 7 5 68 8 5
12 0.01 0.01 0.112 12.53 23.62 0.322 3 12 0.028 24.40 6.024 0.426 5
8 17 73 4 8 18 3 1
12 0.01 0.01 0.094 13.20 21.86 0.269 3 12 0.010 32.16 2.464 0.110 5
1 66 09 9 6 5 6 8
12 0.01 0.01 0.078 13.69 20.66 0.340 3 12 0.042 18.98 11.23 0.656 5
71 07 4 4 99 32 8
12 0.01 0.01 0.100 13.92 20.11 0.240 3 12 0.013 31.58 2.633 0.144 5
4 9 65 5 1 79 9
12 0.01 0.01 0.078 16.09 15.67 0.278 3 12 0.028 24.87 5.704 0.418 5
8 55 56 3 7 58 4 5
12 0.01 0.01 0.090 12.64 23.32 0.282 3 12 0.018 26.51 4.725 0.266 5
7 39 4 7 4 1 5 8
12 0.01 0.01 0.120 13.19 21.87 0.379 3 12 0.038 23.13 6.968 0.567 5
9 92 97 7 5 74 3 1
12 0.01 0.01 0.083 15.18 17.40 0.272 3 12 0.022 26.57 4.691 0.316 5
83 14 3 4 47 3
12 0.01 0.01 0.146 11.25 27.35 0.373 3 12 0.024 26.93 4.500 0.335 5
9 78 95 5 2 42 8 8
12 0.01 0.01 0.102 14.46 18.90 0.412 3 12 0.046 21.35 8.555 0.708 5
8 76 69 3 5 47 9 2
12 0.01 0.01 0.082 15.08 17.60 0.236 3 12 0.018 28.20 3.888 0.231 5
8 73 5 4 3 51 2 9
12 0.01 0.01 0.106 12.68 23.22 0.421 3 12 0.035 22.33 7.645 0.514 5
4 11 44 7 17 7 7
12 0.01 0.01 0.075 15.60 16.58 0.296 3 12 0.018 28.06 3.949 0.252 5
9 77 12 1 1 86 8 6
12 0.01 0.01 0.113 11.94 25.27 0.330 3 12 0.033 22.59 7.418 0.506 5
7 62 5 9 4 37 5 6
12 0.01 0.01 0.153 8.952 35.67 0.448 3 12 0.013 30.11 3.119 0.221 5
5 2 73 1 4 87 4 1
128 6 0.01 0.01 0.078 15.12 17.53 0.207 3 6 0.020 26.91 4.512 0.300 5
4 07 73 2 1 9 4
6 0.01 0.01 0.078 13.80 20.39 0.231 3 6 0.012 29.70 3.272 0.203 5
94 53 1 5 21 6 8
6 0.01 0.01 0.068 15.59 16.60 0.323 3 6 0.020 26.27 4.856 0.315 5
8 5 54 3 1 38 3 1
6 0.01 0.01 0.028 24.39 6.027 0.093 4 6 0.015 29.67 3.281 0.205 5
8 74 4 6 7 74 9 9
6 0.01 0.01 0.008 33.47 2.120 0.074 5 6 0.007 34.43 1.897 0.123 5
2 22 3 1 3 46 9 4
6 0.01 0.01 0.016 29.20 3.466 0.169 5 6 0.016 29.46 3.363 0.194 5
8 23 5 6 3 54 1
6 0.01 0.01 0.028 25.23 5.472 0.272 4 6 0.018 29.15 3.485 0.292 5
3 6 7 9 41 7 4
6 0.01 0.01 0.016 29.45 3.366 0.157 4 6 0.021 27.08 4.422 0.292 5
1 59 7 6 1 75 1 2
6 0.01 0.01 0.016 28.38 3.807 0.050 4 6 0.011 31.54 2.645 0.147 5
4 73 4 7 4 98 5 5
6 0.01 0.01 0.025 23.47 6.7 0.217 4 6 0.026 23.22 6.899 0.456 5
3 85 5 31 9 4
6 0.01 0.01 0.024 26.41 4.778 0.158 4 6 0.021 27.40 4.261 0.285 5
41 5 1 4 86 6 5
6 0.01 0.01 0.014 30.86 2.862 0.075 5 6 0.013 31.46 2.673 0.208 5
3 41 8 1 3 1
6 0.01 0.01 0.014 28.42 3.789 0.102 5 6 0.014 28.76 3.647 0.188 5
7 96 1 2 5 5
6 0.01 0.01 0.036 23.52 6.661 0.288 4 6 0.025 26.57 4.691 0.391 5

Page | 26
 8 91 1 2 9 38 5 4
6 0.01 0.01 0.019 27.63 4.151 0.175 4 6 0.017 28.59 3.715 0.217 5
8 58 5 8 7 94 6 2
6 0.01 0.01 0.022 27.53 4.200 0.078 4 6 0.020 28.48 3.764 0.251 5
6 47 2 3 2 5 9 1
6 0.01 0.01 0.048 21.03 8.875 0.298 4 6 0.030 25.15 5.526 0.497 5
3 58 9 3 1 13 3 4
6 0.01 0.01 0.022 26.33 4.821 0.178 4 6 0.016 29.24 3.449 0.210 5
7 65 4 9 2 5 5 8
6 0.01 0.01 0.038 21.54 8.365 0.268 4 6 0.023 25.77 5.144 0.356 5
3 98 8 7 6 29 7 9
6 0.01 0.01 0.017 28.47 3.771 0.131 5 6 0.014 29.73 3.262 0.181 5
3 07 1 6 9 01 1 3
6 0.01 0.01 0.030 23.32 6.819 0.202 4 6 0.022 26.01 5.003 0.347 5
7 47 7 5 5 54 7
6 0.01 0.01 0.027 23.83 6.428 0.307 4 6 0.013 29.82 3.224 0.193 5
7 72 9 2 9 96 9 5
256 3 0.01 0.01 0.022 26.10 4.95 0.132 4 3 0.014 29.70 3.270 0.255 4
1 79 2 6 74 6 7
3 0.01 0.01 0.021 25.04 5.592 0.096 4 3 0.012 29.74 3.254 0.226 5
4 8 4 4 4 97 7 7
3 0.01 0.01 0.024 24.66 5.844 0.171 4 3 0.014 29.12 3.497 0.235 5
2 46 8 3 5 4 9 4
3 0.01 0.01 0.007 35.62 1.655 0.024 5 3 0.027 24.74 5.789 0.345 5
9 8 9 7 66 9 4
3 0.01 0.01 0.008 33.55 2.101 0.042 3 0.007 34.43 1.897 0.123 5
1 12 1 2 3 46 9 4
3 0.01 0.01 0.011 32.26 2.437 0.098 5 3 0.014 30.50 2.984 0.249 5
8 07 6 4 4 31 3 9
3 0.01 0.01 0.013 31.60 2.630 0.142 5 3 0.022 27.14 4.390 0.419 5
6 04 1 2 7 94 6 7
3 0.01 0.01 0.012 31.98 2.516 0.120 5 3 0.019 27.76 4.090 0.258 5
36 6 6 5 48 3 1
3 0.01 0.01 0.006 36.80 1.445 0.020 5 3 0.015 28.78 3.638 0.206 5
2 14 2 2 7 27 4
3 0.01 0.01 0.013 29.17 3.478 0.146 5 3 0.015 27.64 4.145 0.221 5
1 19 6 7 6 86 4 9
3 0.01 0.01 0.011 32.51 2.366 0.098 5 3 0.018 28.50 3.758 0.289 5
9 78 5 2 9 05 2 3
3 0.01 0.01 0.009 34.67 1.846 0.056 5 3 0.018 28.54 3.737 0.310 5
2 48 1 2 6 77 8 4
3 0.01 0.01 0.012 30.07 3.134 0.063 5 3 0.013 29.01 3.542 0.181 5
2 76 2 6 8 38 5 5
3 0.01 0.01 0.017 30.16 3.102 0.170 5 3 0.019 28.98 3.556 0.346 5
1 47 9 9 7 03 2 6
3 0.01 0.01 0.011 32.25 2.440 0.087 5 3 0.015 29.80 3.233 0.257 5
6 01 6 8 4 79 3
3 0.01 0.01 0.009 35.36 1.705 0.042 5 3 0.021 28.03 3.963 0.298 5
2 09 9 5 3 84 5 7
3 0.01 0.01 0.018 29.30 3.426 0.178 5 3 0.023 27.23 4.346 0.307 5
6 4 1 5 6 79 2 8
3 0.01 0.01 0.010 32.83 2.281 0.109 5 3 0.014 30.01 3.156 0.267 5
7 7 1 5 8 56 6 9
3 0.01 0.01 0.018 27.99 3.981 0.151 5 3 0.019 27.45 4.237 0.309 5
2 87 7 1 4 76 6 6
3 0.01 0.01 0.011 31.77 2.578 0.072 5 3 0.013 30.34 3.040 0.224 5
8 39 1 5 9 2 2 6
3 0.01 0.01 0.016 28.80 3.627 0.155 5 3 0.019 27.36 4.283 0.243 5
3 85 2 9 3 31 9 5
3 0.01 0.01 0.014 29.58 3.318 0.168 5 3 0.014 29.40 3.388 0.279 5
3 17 3 6 02 4

Page | 27
 𝒅 𝒆

Fig 6(d)and(e): Illustrates the comparison of DCT compression method and CS method for signal taken
from the MIMIC database with M=256

In the figure 6(d) and (e) shows the comparison between the 2 methods, ie. DCT
compression method and CS based method for a sample of length N=512 and measurement
M=256 which is taken from the MIMIC data base. The measurements and error signal for
the original signal by both the methods were shown in figure 6 (d). The reconstructed
signals using the both the methods along with the original signal is illustrated in figure 6 (e).
By analysing the reconstructed waveform from the figure it is seen that waveform
generated by method 2 ie, DCT compression based method shows better quality .

Page | 28
 7 Conclusion

Compressive sensing has emerged as an area that opens new perspectives in signal
acquisition and processing. Investigated the performances of CS based data reduction
algorithm for different combination of representation techniques. In this work, we consider
different types of signals for evaluating the compression performance of the proposed
method. The signal of length N=512 from the MIMIC database with sampling rates Fs = 125
Hz, slp database with sampling frequency Fs = 250 Hz, ourdatabase with sampling frequency
Fs = 250 Hz to estimate ̅ . Select the signals from the database and resampled it to 250 Hz.
For quantization the bits per measurement set as BM=10 bits. For 512 samples the
compressive measurements taken as M=64,128 and 256. The proposed method is
implemented using MATLAB software. The CR and PRD metrics are used for assessing the
quality of the reconstructed signal. From the results it is observed that the CS method had
higher Compression Ratio with PRD value of below 10 clearly indicate that the proposed
basis matrix Ψ= *I C S+ is capable of reproducing the original. CS based methods were also
implemented in Arduino computing platform for performance analysis.The CS based
method is compared with the DCT based PPG compression algorithm. From that the quality
of the reconstructed waveform is analysed, DCT based compression algorithm shows better
results.

Page | 29
 Reference

1. M. Sabarimalai Manikandan et al., " Simultaneous denoising and compression of power

system disturbances using sparse representation on overcomplete hybrid dictionaries," IET
Gener. Transm. Distrib., 2015, Vol. 9, Iss. 11, pp. 1077–1088
2. P. Kathirvel et al., " Detection of Power Quality Disturbances with Overcomplete Dictionary
Matrix and ℓ1-norm Minimization," 2019 IEEE 16th International Conference on Wearable
and Implantable Body Sensor Networks (BSN), Chicago, IL, USA, 2019, pp. 1-4.
3. C. Park and B. Lee, "Energy-Efficient Photoplethysmogram Compression to Estimate Heart
and Respiratory Rates Simultaneously," in IEEE Access, vol. 7, pp. 71072-71078, 2019.
4. H. Zamani, F. Marefat and P. Mohseni, "Block-Sparse Compressive Sensing for High-Fidelity
Recording of Photoplethysmogram," 2018 IEEE Biomedical Circuits and Systems Conference
(BioCAS), Cleveland, OH, 2018, pp. 1-4.
5. C. Park and B. Lee, "Compressed estimation of heart and respiratory rates from a
photoplethysmogram," 2017 IEEE Biomedical Circuits and Systems Conference (BioCAS),
Turin, 2017, pp. 1-4.
6. V. Natarajan and A. Vyas, "Power efficient compressive sensing for continuous monitoring of
ECG and PPG in a wearable system," 2016 IEEE 3rd World Forum on Internet of Things (WF-
IoT), Reston, VA, 2016, pp. 336-341.
7. J. Luan, S. Lee and P. H. Chou, "Fast Compressive Sensing Based on Dominant Frequency
Estimation," 2015 IEEE 12th International Conference on Mobile Ad Hoc and Sensor Systems,
Dallas, TX, 2015, pp. 454-455.
8. Xiao J, " A Low-Complexity Compressed Sensing Reconstruction Method for Heart Signal
Biometric Recognition," in Sensors (Basel)., vol. 62, pp. 19(23):5330, Dec. 2019.,
9. Z. Zhang, "Heart rate monitoring from wrist-type photoplethysmographic (PPG) signals
during intensive physical exercise," 2014 IEEE Global Conference on Signal and Information
Processing (GlobalSIP), Atlanta, GA, 2014, pp. 698-702.
10. P. K. Baheti and H. Garudadri, "An Ultra Low Power Pulse Oximeter Sensor Based on
Compressed Sensing," 2009 Sixth International Workshop on Wearable and Implantable
Body Sensor Networks, Berkeley, CA, 2009, pp. 144-148.

Page | 30
11. Candes, E.J.; Tao, T. Near-optimal signal recovery from random projections: Universal
encoding strategies. IEEE Trans. Inf. Theory 2007, 52, 5406–5425.
12. Donoho, D.L. Compressed sensing. IEEE Trans. Inf. Theory 2006, 52, 1289–1306.
13. Baheti, P.K.; Garudadri, H.; Majumdar, S. Blood oxygen estimation from compressively
sensed photoplethysmography. In Proceedings of the Wireless Health 2010, WH 2010, San
Diego, CA, USA, 5–7 October 2010; pp. 10–14.
14. Lee, S.; Luan, J.; Chou, P. A New Approach to Compressing ECG Signals with Trained
Overcomplete Dictionary. In Proceedings of the 2014 4th International Conference on
Wireless Mobile Communication and Healthcare—Transforming Healthcare Through
Innovations in Mobile and Wireless Technologies (MOBIHEALTH), Athens, Greece, 3–5
November 2014; pp. 83–86.
15. Irena Orovi, et al.,” Compressive Sensing in Signal Processing: Algorithms and
Transform Domain Formulations” Hindawi Publishing Corporation Mathematical
Problems in Engineering, Volume 2016, Article ID 7616393, 16 pages

Page | 31
 Acknowledgement

We are very much thankful to Ministry of Electronics and Information Technology (MeitY)
government of India, for initiating a futuristic program, which is Special Manpower
Development Programme in VLSI Design, development of Related Software and design of
final prototype by using ASICs. Special Manpower Development Program for Chips to
System Design (SMDP-C2SD) has sole purpose to increase India's share of the Global VLSI
Design market by making available specialized manpower required by the Design Centers/
Industry.
Also we are thankful to IIT Bhubaneswar for channelizing such challenging research activity.
Our sincere thanks to Dr. M S Manikandan for his esteemed guidance and support, without
his vision this work might not be possible.

Page | 32